mind the gap! roman republican coin hoards from italy and iberia … · 2018. 2. 21. · republican...
TRANSCRIPT
Mind the Gap! Roman Republican coin hoards from Italy
and Iberia at the end of the second century BC
Kris Lockyear
1 Introduction
For many years I have been both collecting and analysing data on Roman Republican coin
hoards.1 The aim of these analyses has been to identify patterns of coinage found within
these hoards, and to understand how these patterns came about. We can, for example, see
how the coins in hoards that close immediately after the Battle of Thapsus in 46 BC reflect
where Caesar’s troops, and those of the optimates, went to after the battle.2 During these
analyses, I observed an unusual pattern in hoard closing dates for the period 118–97 BC.3
From 118–113 BC the majority of hoards came from Italy or Sicily, after which there is a ten-
year gap in the Italian sequence until about 104 BC. This gap is, in itself, highly unusual in a
period of continuous coin manufacture. What makes this gap even more enigmatic is the fact
that in the same period there is a peak in hoarding activity in the Iberian Peninsula. This peak
in the hoards from Iberia exceeds all other periods during the Republic, even the Civil Wars.
At the time my interpretations were tentative, as the analyses were based on Crawford’s
dating scheme4 (see below). In 2007 I returned to this problem and analysed an expanded
data set by individual issues.5 This revealed that the pattern persisted strongly at the level of
individual issues and that certain issues appeared to be key points in the sequence, especially,
for example, RRC 286/1.6
This pattern, which for the sake of brevity I refer to as ‘the gap’, raises a number of
questions. Is ‘the gap’ real, or is it a function of our dating schemes? If it does exist, can we
identify which issues are involved? Why does this gap exist, and what might it tell us about
the nature of coinage supply and use in the later Roman Republic?
The supply of Roman Republican coinage to the Iberian Peninsula was episodic, with the
peak of coinage supply in the late second century noted above, and further peaks in the 70s
BC and in the 40s BC during the first decade of the Civil wars (see Fig. 17).7 Despite this
episodic supply over the two centuries from the Second Punic War to the end of the Republic,
Roman issues became the principal form of coinage in the Iberian Peninsula. Also, during
these two centuries, many local coinages were struck on the Roman model, the so-called
1 Lockyear 2007, 2016.
2 Lockyear 2007, pp. 116–20.
3 Lockyear 1996b, sections 8.3.3–8.3.4.
4 Crawford 1974.
5 Lockyear 2007, pp. 77–9.
6 Abbrevations used in this paper: RRC = Roman Republican Coinage (Crawford 1974); RRCH = Roman
Republican Coin Hoards (Crawford 1969c); IGCH = An Inventory of Greek Coin Hoards (Thompson et al.
1973).
7 Lockyear 2007.
‘Iberian denarii’, the date and function of which is still much debated.8 Useful summaries of
the monetary history of the area have been published by Ripollés.9
This paper has two aims. Firstly, it seeks to re-examine ‘the gap’ in depth and assess the
strength and nature of the pattern, including which issues are associated with it. Secondly, it
seeks to assess the impact on that pattern of the proposed re-dating of the issues from the end
of the second century and the beginning of the first century BC by the late Harold
Mattingly.10 In working to achieve these two principal aims, the paper also demonstrates both
the advantages and limitations of multivariate statistical analysis of coin hoard data. To
conclude, some possible explanations for the pattern are discussed.
One may question why such an array of multivariate statistical techniques have been applied
to this problem. Why not simply list the hoards and issues associated with ‘the gap’? Such a
simplistic approach would be reliant on our available chronologies, be they Crawford’s,
Mattingly’s or any other scheme. With such a subtle pattern, we need to apply techniques
which are independent of those dating schemes. The difficulties are compounded by two
other aspects of the data. Firstly, the issues vary greatly in size, some are very common,
some are very rare. It is likely that part of the patterning we can observe in the data may be
the result of simple random variation. We need methods which give us some indication of
the confidence we can place on the results, not just in overall terms but also on each
individual hoard and/or issue. Secondly, coins circulate. Just because an issue is released in
place A, does not mean that in time it will not form part of the coinage pool in place B. For
the current problem, issues released into the coinage pool in Italy will eventually arrive in
Iberia and vice versa. This will inevitably cause some blurring of any initial pattern.
1.1 Dating schemes
The dating scheme presented in Roman Republican Coinage is not universally accepted, in
part because of the tight date ranges offered that seem to many to be overly precise given the
evidence available, and in part because new finds have forced a reassessment of some parts of
the scheme. For the period under consideration here, the break in Crawford’s sequence
between the issues of 100 BC and 92 BC is particularly controversial, especially as eight
issues are squeezed into two years from 101–100 BC.
The most recent attempt to re-date these issues is that by Mattingly.11 The new scheme is
based on a combination of hoards and literary evidence and is presented in a series of three
tables. Only the last of those tables, however, provides fixed dates for the issues. The
remaining tables only provide a relative ordering, and so I have worked back through them
assigning dates sequentially.12 Many of the analyses below are independent of the dating
schemes, but my imposition of dates on Mattingly’s tables should be kept in mind.13 One of
8 Crawford 1985; DeRose Evans 2013; Knapp 1987.
9 Ripollés 2005, 2012.
10 Mattingly 2004, chapter 13.
11 Ibid.
12 RRC 324 occurs twice in Table 2 on p. 206. The first occurrence is incorrect and should be RRC 321 which
is re-dated to 100 BC. RRC 211–13, 272 and 339 were omitted from Mattingly’s scheme as they consist only of
bronzes.
13 In the Coin Hoards of the Roman Republic database (CHRR; Lockyear 2007, chapter 2) I adopted a crude
but effective method of implementing these alternate dating schemes by simply duplicating the COINTYPE table
and editing the affected issues. This enabled comparative data sets for analysis to be created with ease. One
impact of simply assigning dates back sequentially through Mattingly’s tables is that the earliest issues
Mattingly’s aims was to assign issues to the period 99–93 BC, left almost devoid of denarius
issues by Crawford.
The impact of Mattingly’s changes can be assessed by looking at the impact of the change
in the dates. Of the 144 issues affected, only ten issues have a date change of five or more
years and 77 issues change by two or less years. Looking at the closing dates14 of the 71
hoards used in this paper,15 the biggest change is seven years, which affects only three
hoards, whereas 49 hoards have a change of less than four years. In terms of broad-brush
dating of the issues it would seem that the new scheme has a relatively minor impact.
1.2 The data
Since the publication of Patterns and Process16 efforts have been made to improve the
coverage in the CHRR database for material relating to this period. In particular, listings for
the hoards used by Mattingly were sought out and entered. The data are not, however,
without problems. For example, the Terranova de Sicilia and Manfria hoards (TDS & MNF) are
argued by Crawford to be two lots of a single hoard17, whereas the La Oliva hoard (RRCH
197) is thought to be two hoards by Chaves Tristán.18 One must also remember that not all
the hoards have the best data quality. For example, the La Oliva hoard (OL2) has only 45
coins recorded from an estimated 600–700. The São João dos Caldeireiros hoard (SJ1) is a
particularly complex hoard. This hoard — also known as the Mértola hoard — is
unpublished, although it has been examined by both Michael Crawford and Terrence Volk.19
My listing is derived from Crawford’s lists now housed in the British Museum.20 Mattingly’s
list, created from a combination of Volk’s and Crawford’s, varies from the data used here,
however. For example, Mattingly has 15 examples of 286/1, M. SERGI. SILVS Q., whereas
Crawford lists 26 examples.
Rather than weed the data set of all potentially problematic data — which would almost
certainly leave us with no data at all — I prefer to work through the analyses and omit any
problematic assemblages as they arise, noting, of course, which these were and why.
Appendix A contains details of all the hoards used, whilst Table 1 provides a summary.
[insert Table 1 near here]
Included in the analysis in 2007 were hoards closing between 118 and 97 BC. However,
use of Mattingly’s dates changes the hoard selection slightly, for example the Patrica and La
Loba hoards (PTA & LOB) date to 119 BC using Crawford’s dates and 118 BC using
Mattingly’s. Using a date bracket of 119–97 BC (RRC dates)21 gives us 71 hoards which
contain at least one denarius (Table 1, and Appendix A). Of those hoards, 18 have less than
considered by Mattingly (his Table 4, RRC 197 onwards) are dated from 153 BC to 133 BC. This creates an
artificial gap between the issues prior to RRC 197 which Crawford dated to 169–158 BC and RRC 197. For our
purposes here, however, this problem is irrelevant.
14 Where the latest coin in a hoard is assigned a date range, the earliest date is used.
15 Previous analyses have included the hoard from Borgonuovo (BRG) due to an error in the database.
16 Lockyear 2007.
17 Crawford 1974, note to table XI, page 67.
18 Chaves Tristán 1996, Nos. 24 & 32, here hoards OLI & OL2.
19 Mattingly 2004, notes 3 and 7, pp. 200–1.
20 I am extremely grateful to Michael Crawford for helping me with these. They are less clear than usual due to
the extreme haste with which the list was created.
21 The four hoards with a closing date of 97 BC derive that date from quinarii.
30 well identified denarii22— the minimum hoard size used previously — leaving us with 53
hoards compared to the 41 hoards used in Patterns and Process. Some of the very small
hoards clearly have an unreliable closing date. This is nicely illustrated by the mixed
denomination Panicale hoard (PAN) which closes in 89 BC with two quinarii of RRC 343, but
the latest denarius dates to 115 BC. Similarly, the Bugiuleşti hoard (BUG), that has only two
coins and closes in 112 BC, cannot have been deposited much before the massive export of
denarii to Dacia in c.75 BC.23
All the hoards bar two come from Italy/Sicily or Iberia, the exceptions being the
Bugiuleşti and Lauterach (LAU) hoards which come from Romania and Austria.
2 The analyses
In the following section, I have divided the analyses up into a series of questions and
answers. As much as possible I have tried to describe the methods and the results in plain
English. Technical details have been relegated to Appendix B, and detailed comments on the
data or the techniques have been relegated to footnotes. I have not included any formal
mathematical justification for the methods used as they are all standard techniques widely
discussed in the statistical literature.
Multivariate statistical techniques can be seen as simultaneously simple and complex.24
They are complex in the sense that the underlying mathematics is non-trivial. They are
simple, however, in that with modern software, undertaking the analyses is relatively
straightforward. Provided the analyst understands the aim of the technique, how to interpret
the results, and its limitations, it is possible to apply these methods without a detailed
understanding of the underlying mathematics.
2.1 How has the new data, and Mattingly’s dating scheme, affected ‘the gap’?
Figure 1 shows the closing date of the 61 hoards in the data set that have over ten coins, using
Crawford’s dates. ‘The gap’ can be clearly seen: only one Italian hoard closes between 115–
105 BC. Comparing this graph to one calculated using Mattingly’s scheme (Fig. 2) shows a
few differences. For example, the peak at the end is more spread out, reflecting Mattingly’s
attempt to assign issues to 99–93 BC. There is, however, still only one Italian hoard closing
in ‘the gap’, which has the same duration but with slightly different beginning and end dates.
Figure 1: Closing date of hoards using the dating scheme from RRC. The ‘gap’ lies
between the dashed lines.
Figure 2: Closing date of hoards using the dating scheme from Mattingly 2004. The
‘gap’ lies between the dashed lines.
22 A “well identified” denarius is one which can be given a reference down to a single RRC issue if not to an
exact type. For example, a coin which can be said to be RRC 299, but not specified if it is 299/1a or /1b is
considered “well identified” and is given the appropriate accuracy code in the database (see Lockyear 2007, pp.
17–19). In Table 1, the total number of well identified coins is given in the column “good total”.
23 Lockyear 1995, 1996a, 2008.
24 Baxter 2003, pp. 16–18.
2.2 Can ‘the gap’ in the closing dates of hoards be seen in the overall structure of the
hoards?
Just looking at closing dates does not, obviously, take into account the hoard structure, i.e.,
the quantities of coins of different dates within each hoard. This can be very informative. For
example, Republican hoards from Dacia have been shown to have a very similar structure,
whatever their closing date.25 That structure closely resembles Italian hoards of the mid 70s
BC, indicating that large numbers of denarii entered Dacia at about that date, after which the
supply was much lower and somewhat erratic. One way to examine hoard structure is to use a
technique called Hierarchical Cluster Analysis. This technique aims to divide the
assemblages, whatever they are, into groups or ‘clusters’ on the basis of some form of
measurement, in this case the numbers of coins by year of striking.26 The analyst may be
looking for ‘real’ groups in the data, or may be seeking to simply subdivide (or ‘dissect’) a
continuum of variation in a repeatable fashion. In the case of coin assemblages, we are
unlikely to have firm groups and are normally dissecting a continuum.
There are many varieties of Hierarchical Cluster Analysis and the one chosen for the
current analysis — Dmax-based cluster analysis — has proved successful in the analysis of
coin data previously and has been explained in detail elsewhere.27 It has been used
successfully in the analysis of both hoards28 and site finds.29 The technique has two stages:
Figure 3: Calculating the Kolmogorov-Smirnov distance, Dmax, for three hoards
closing in 100 BC.
1. Firstly, a measure of how similar (or dissimilar) the hoards are to each other has to be
calculated. Many such measures are available but here we will use the Kolmogorov-
Smirnov distance, Dmax, as a measure of dissimilarity. This is most easily explained
via a graph (see Fig. 3). Each hoard is converted into a cumulative proportion curve,
i.e., the number of coins for each year expressed as a proportion, is added to the sum
of all the previous years to create a curve that goes up from zero to one. Each curve is
then compared to every other curve and the maximum absolute difference is our
measure of dissimilarity. In Fig. 3 the curves for Paterno (PAT) and Bevagna (BVG)
show that the two hoards are very similar to each other. Consequently, the value of
Dmax is small: 0.08. Conversely, hoards Olival da Soalheira (OSO) and Paterno (PAT)
are very different and thus Dmax is large, 0.52.
2. Secondly, the hoards are now iteratively joined into groups called clusters. Firstly, the
two most similar hoards, i.e., the pair with the smallest value of Dmax, are joined into
a cluster and the average dissimilarity between that cluster and the rest of the hoards
is calculated. Having done that, the next smallest value for Dmax is found, the hoards
joined into a cluster, and the Dmax values recalculated. The process continues until
all the hoards form one large cluster. This technique, known as average linkage, is
25 Lockyear 1995, 1996a, 2008.
26 See, for example, Baxter 1994, chapter 7.
27 e.g., Lockyear 2000.
28 Lockyear 1995, 1996a.
29 e.g., Walton 2012.
only one of a number of methods for creating clusters.30 It was chosen because it was
successful in the previous analyses. The results can be presented as a tree-like graph
called a dendrogram. The lines in the dendrogram show at which level of dissimilarity
each hoard is joined to a group. The dendrogram can be ‘cut’ to create the desired
number of clusters and shows which hoards are in each one.
The analysis was run twice: once using Crawford’s dates and once using Mattingly’s. As the
technique uses cumulative proportion curves, very small hoards are likely to be a problem. In
the past I have used a minimum hoard size of 30 denarii.31 Running the cluster analysis using
the 53 hoards of this size suggested five groups in the data: two ‘singletons’ (i.e., clusters
with only a single member, in this case hoards CO1 and PZ1) and three larger clusters
corresponding to an early, middle and late group (see Figs. 4– 5).
Figure 4: Dendrogram from cluster analysis, RRC dates. Iberian Peninsula hoards in
italics.
Figure 5: Dendrogram from cluster analysis, Mattingly’s dates. Iberian Peninsula
hoards in italics.
The early cluster contains 15 hoards with a median closing date of 114 BC (RRC) or 111
BC (Mattingly). Eight of the hoards come from Italy and seven from Iberia. It represents,
therefore, early hoards (or hoards with an ‘archaic’ structure) and is evenly split between the
two regions.
The middle cluster from both analyses contains 23 hoards in common. Two hoards,
Almadenejos and Idanha-a-Velha (ALM & IAV), occur in the middle group using Crawford’s
chronology and the late group if using Mattingly’s. The middle group has a median closing
date of 103 BC (RRC) or 101 BC (Mattingly). Of the 23 hoards in common, five come from
Italy and eighteen come from Iberia. Both the two extra hoards in this group when using
Crawford’s chronology also come from Iberia. This cluster shows a clear bias to Iberia and
mainly includes hoards from ‘the gap.’
The late cluster contains 11 hoards in common, with the addition of the two hoards
mentioned above when using Mattingly’s chronology. The cluster has a median closing date
of 100 BC (RRC) or 95 BC (Mattingly). In this case nine of the hoards come from Italy and
only two or four from Iberia, depending on the dating scheme used.
Of the two singletons, Pozoblanco (PZ1) is unusual in having a large percentage of coins
of issues RRC 281 and 282,32 the latter being the issue struck in Narbo in 118 BC. The other
singleton, Villanueva de Córdoba (CO1), is unusual in that 40% of that hoard consists of
issues from RRC 274–281, giving it a very modern profile.33
30 Shennan 1997, pp. 239–40.
31 Lockyear 2007, p. 183.
32 38%, Lockyear 2007, p. 72
33 See Lockyear 2007, Fig. 5.20, p. 70.
This analysis has clearly shown that the pattern previously noted goes deeper than just the
closing date of the hoards, but is also a reflection of the overall coin hoard structure. Figure 6
shows the average cumulative percentage curve from each cluster. The early cluster contains
both hoards with early closing dates, but also hoards with a later closing date but an archaic
structure such as Cogollos de Guadix, Puebla de los Infantes and Cástulo (COG, PUE & CSL).
Similarly, the middle cluster also contains some late closing hoards such as Olmeneta, Olival
da Soalheira and São João dos Caldeireiros (OLM, OSO & SJ1). The late cluster consists almost
exclusively of hoards closing in 100 BC (RRC dates) or 100–94 BC (Mattingly dates).
Figure 6: The average profile of hoards from the three main clusters. Solid lines: RRC dates;
dashed lines: Mattingly’s dates.
2.3 Can we see ‘the gap’ if we analyse the hoards independently of the rival dating
schemes?
The analyses performed so far are reliant on the dating schemes given by RRC or Mattingly.
We can, however, analyse the hoards at the level of individual issues using Correspondence
Analysis.34 This method seeks to identify the underlying trends in the data and to represent
those patterns in a series of graphs, which are accompanied by some diagnostic statistics. It
is, essentially, looking for the pattern whilst trying to ignore the noise. The technique is used
widely in the social sciences35 and has been successfully used in archaeology.36 Formal
discussions of the technique for archaeologists are available,37 and I have provided a detailed
explanation along with worked examples specifically for numismatists,38 as well as
discussions of the technique in relation to the analysis of site find assemblages and hoards.39
In numismatics, the technique has been both criticised and promoted in the same review.40
Readers wishing the more technical details are referred to the citations in the notes. For
our purposes here we simply need to note the following.
1. The aim of this technique is to enable a complex data set to be plotted as a series of
simple two-dimensional graphs. To do this, the technique calculates a series of new
synthetic axes where axis 1 represents the main underlying pattern in the data, axis 2
presents the next most important source of variation, and so on. In coinage studies, the
first axis often represents date (e.g., early to late hoards).
2. The results of the technique are plotted on one or more scattergrams, technically
called maps as both the x and y axes should be at the same scale. Usually two maps
are produced, both showing axes 1 v. 2 but one representing the ‘objects’ (in our case
hoards) and one representing the variables (in our case RRC issues). More subtle
patterns can be observed by examining axes 3 and beyond although in hoard studies
34 Greenacre 2007; Lockyear 2007.
35 e.g., Greenacre & Blasius 1994.
36 e.g., Cool & Baxter 1995, 1999, 2002; Pitts & Perring 2006.
37 Baxter 1994, 2003; Baxter & Cool 2010; Shennan 1997.
38 Lockyear, forthcoming. A draft is available on Academia.edu.
39 Lockyear 2000, 2007.
40 Esty 2003, p. 924, cf. p. 925–6.
these can quickly start representing the variation in only a very small number of
issues.
3. Two points on a map plotted close together are likely to be similar in some way, and
two points plotted at a distance are likely to be dissimilar. In our case here, two hoards
plotted close together are likely to contain a similar range and proportion of coinage
issues. Similarly, two issues plotted close together are likely to have a similar
distribution across the hoards.
4. As the process of calculating the axes and plotting the scores is an attempt to simplify
a complex data set, some items will not fit that simplified pattern well. Consultation
of the accompanying diagnostic statistics (more properly known as decompositions of
inertia) enables one to identify which items (hoards and/or issues) fit the pattern, and
which should be ignored.41 One of the most useful of the diagnostic statistics is
‘quality’ which scores out of 1,000 how well a point ‘fits’ the map. A simple analogy
may help to explain this. Imagine we have a map of Midtown Manhattan on to which
we plot the location of every person at midday on one specific day. We will,
obviously, have lots of points along the sidewalks, and in rows along the roads. We
will also have a huge mass of points at 350 5th Avenue. This mass of points
represents the large number of people in the Empire State Building. For the people on
the sidewalk, the point on the map is a good representation of their location, and in
CA terms would have a high ‘quality’. For the people on the 86th floor observation
deck of the Empire State Building, their coordinates may be accurate in terms of
eastings and northings, but fails to show their height above street level, and would,
therefore, have a low ‘quality’. As a result, the position of items with a very low
quality should not be given any great meaning when interpreting a map from CA. The
diagnostic statistics for the analysis shown here are given in Tables 3 and 4 in
Appendix B.
5. By comparing the maps and consulting the diagnostic statistics, an interpretation for
each axis can be given.
Although the strength of CA is its ability to examine large numbers of hoards and issues
simultaneously and represent them in a small number of graphs, its main weakness is that the
results can be negatively affected by very unusual assemblages.42 The normal procedure is to
note the unusual assemblages and re-run the analysis omitting them.
In my previous analyses the minimum hoard size has been set at 30 denarii. In this
analysis, however, we are only interested in the pattern of the latest issues and so all those
prior to RRC 260 were removed from the analysis, i.e., those issues struck before 128 BC
(RRC). As a result eleven of the hoards fell below 30 coin limit43, but these were retained for
the initial run of the analysis. A further complication arises when one considers Mattingly’s
arrangement. Issues RRC 261–7 included in this data set are placed by Mattingly earlier than
RRC 260. Unfortunately, there is no simple break in the sequence which would result in
identical data sets whichever arrangement was followed, and so I stuck to using the same
issues as we have used so far.
41 I have provided a fuller discussion of the decomposition of inertia elsewhere (Lockyear 2007, 57–9).
42 For example, the coin assemblage from the excavations of Fishbourne Roman Palace had a huge negative
impact on the results of the CA of site find data from Britain (Lockyear 2000).
43 In addition, RRC 309 was removed as there was only a single example in the data set from the Sierra Morena
(El Centenillo) hoard (CEN).
An initial analysis showed that, as previously,44 the Idanha-a-Velha hoard (IAV) has a
strong influence on the results because it has 145 examples of RRC 319, some 70% of the
coins of that type in the data set.45 It was decided to omit that hoard from the rest of the
analyses.
Figure 7: CA map of the coin issues, first (x) and second (y) axes of inertia. Symbols
represent date brackets according to RRC, numbers are RRC issues.
Figure 8: CA map of the coin issues, first (x) and second (y) axes of inertia. Symbols
represent date brackets according to Mattingly, numbers are RRC issues.
Figure 9: CA map of the hoards, first (x) and second (y) axes of inertia. Symbols
represent country of origin. The three letter hoard identifiers are given in Table 1.
Figure 10: CA map of the hoards, first (x) and second (y) axes of inertia. Symbols
represent clusters from the previous analysis. The three letter hoard identifiers are
given in Table 1.
Figures 7–10 show the results from the CA of the final data set. Figure 7 shows the map
of the issues. The first axis has early issues plotted to the left and later issues plotted to the
right, i.e., the classic time gradient seen in many CAs. The second axis has middle period
issues plotted at the bottom, and the later issues plotted at the top. As a result, the issues lie in
a classic ‘horseshoe curve’ on the map, a very common result in data sets with a strong
underlying gradient such as time, space, or even social status. The symbols in Figure 7
represent periods of time according to the RRC scheme which can be compared to Figure 8
where the same results are presented using Mattingly’s dating scheme. Although varying in
detail, the two schemes are remarkably in agreement in their broad brush strokes. RRC issues
280–286, dating to 120–116 BC, are the least coherent group in Fig. 7 although RRC 283 and
284 both have a low quality on this map and their position should be taken with caution (see
Table 3). On the evidence of the maps, at least, Crawford’s grouping of the issues from 105–
100 BC seems more in keeping with the hoard evidence than Mattingly’s, although as we
shall see, translating the CA maps into precise chronological sequences is not without
difficulties. Apart from issues RRC 271 and 286, there is also a noticeable gap between the
batch of early issues in the upper left quadrant and the batch of issues in the lower right. RRC
271 is quite a rare issue with only 20 coins in the data set, and a low quality of 22; its precise
position, therefore, is not meaningful. RRC 286, however, is a common issue with 234 coins
in the data set and a quality of 316, and therefore its position between the two batches of
issues is meaningful.
Figures 9–10 are maps of the hoards themselves, the former with symbols representing
where they were found and the latter the clusters according to the analysis conducted above
(section 2.2). Unsurprisingly, the clusters do indeed arrange themselves quite nicely along
the first axis with a little mixing at either extreme. The Manfria hoard (MNF) is somewhat
oddly placed lying with the early hoards despite a closing date of 103 BC. It has a lowish
quality on this map of 124 (Table 4). This small hoard of 33 coins ‘out of about 100’46 has a
large gap between RRC 295/1 and its last coin, RRC 320/1. Crawford47 believes that this
44 Lockyear 2007, p. 77.
45 cf. Volk 1994, pp. 47–8, 51.
46 Crawford 1969c, No. 198.
47 Crawford 1974, note to Table XI, p. 67.
hoard is part of the Terranova di Sicilia hoard (TDS) and states that the published list is
incorrect and the hoard actually ends with the issue of M. Aurelius Scaurus, RRC 282/1.48 As
can be seen, in terms of its overall profile it does indeed look like an early hoard as its
position on the map lies close to the other early hoards. This illustrates how robust the
technique can be even when there are a few potential errors in the data.
In terms of geographical origin, Figure 9 shows the hoards in the upper left quadrant are
mainly Italian, those in the bottom left are mainly Iberian, and those to the right are mainly
Italian once more. The pattern seen in the basic closing dates of the hoards (Figs. 1–2)
persists through into this analysis by individual issues. Any suspicion, therefore, that the
pattern is merely an artefact of the dating schemes used, can be dispelled. Some of the
hoards, especially those lying near the origin of the graph, have a very low quality. Often
these are very small hoards, e.g., Aznalcóllar (AZN) has a quality of 14 and only 20 coins, and
Olival da Soalheira (OSO) has a quality of 18 although it is slightly larger with 45 coins. The
position of these hoards on this map cannot be given any weight. Conversely, many hoards
have a high quality, e.g., Imola (IMO, 798), Maddaloni (MAD, 700) and La Oliva (OLI, 581)
and their positions can be treated with confidence.
Comparing Figures 7 and 9 we can see that there is group of Iberian hoards, mainly
belonging to the middle cluster, plotted in the bottom left quadrant, in precisely the area
where there is a gap in the sequence of issues. We have to interpret this in terms of the axes.
The first axis contrasts early issues (approximately RRC 260–283) with late issues
(approximately RRC 311–329). The second axis picks out the middle issues (approximately
RRC 285–306). The Iberian hoards in the lower left quadrant, therefore, have a mix of early
and middle period coin issues, the hoards in the upper left quadrant mainly early coins, and
those in the upper right quadrant mainly later coins. We can illustrate this by using
proportional circles to represent the abundance of each coin type within a hoard. Fig 11
shows a selection of issues as proportional circles plotted on the CA map.
Figure 11: The abundance of four RRC issues in hoards plotted on the CA map. Small open
circles are hoards that have no coins of that issue, hoards plotted in light gray have a low
quality (< 100). The area of the filled circles is proportional to the percentage that issue
forms within a hoard. For hoard names, cf. Fig. 9.
There is a great deal more which could be obtained from a detailed examination of the
results of this analysis, but for our purposes we have achieved our objective of showing that
‘the gap’ can be seen in the hoard data at the level of the individual issues.
2.4 How sure can we be of the position of the hoards and issues in the CA? Can we use
them to suggest a new ordering of issues?
The hoards we are using are, we assume, random selections of denarii in circulation.
Obviously, the distribution of coin types in a small hoard is going to be a less reliable guide
to the coinage pool than the distribution within a moderate sized hoard. Likewise, rare issues
will occur almost randomly in hoards and are therefore less easy to date. As numismatists, we
are instinctively aware that this is the case.
48 Unfortunately, I have been unable to locate Crawford’s revised listing of the Manfria hoard in his archive and
so, for purposes of illustration, have used the published list here.
CA, however, is a purely geometrical technique. It takes the data, in this case coins and
hoards, and produces information about the underlying pattern. Can we investigate how ‘real’
the pattern is, and how much the results might vary simply by chance? Bootstrapping is a
statistical technique which allows the analyst to examine such problems. In bootstrapped CA
the data is resampled to produce, essentially, a simulated set of hoards on which CA is
performed. By resampling and reanalysing the data repeatedly, usually around 10,000 times,
we can visually assess how much an individual hoard or issue would vary simply due to
chance variation.49
The results of a bootstrapped CA are presented in Figs. 12–13 where each ellipse contains
c. 95% of all the points from the 10,000 replications, whilst the point represents the actual
coin issue or hoard.50 A small ellipse indicates a point, either a hoard or an issue, where its
position did not vary much between analyses and one in which we can be confident of the
position of that point. A large ellipse indicates a point where there are large differences
between the analyses and where we should be cautious about placing too much weight on its
position when interpreting the results. In Fig. 12 I have highlighted the ellipses for the twenty
‘best’ hoards, i.e., those with the smallest ellipses and which, therefore, did not vary much
when resampled. Many of these are the larger hoards, such as the São João dos Caldeireiros
hoard (SJ1) which has 464 coins in the analysis, although some are smaller, e.g., the
Maddaloni hoard (MAD) which has 117 coins in the analysis. We can also see the hoards
which are more problematic. In Fig. 12 I have highlighted the Manfria hoard, discussed
above, and the large size of the ellipse clearly shows that this small hoard with only 17 coins
in the analysis is unreliable.51 This is reflected in the quality for these hoards mentioned
above: 124 for Manfria, 798 for Imola and 700 for Maddaloni.
Figure 12: Bootstrapped CA map of the hoards, first and second axes of inertia. The twenty
smallest ellipses have been plotted in black. Manfria’s ellipse is indicated by the broken
line, see text for details.
Figure 13: Bootstrapped CA map of the issues, first and second axes of inertia. The twenty-
three smallest ellipses have been plotted in black. The ellipse for RRC 271 is indicated by
the broken line, see text for details.
Of perhaps more interest is the map for the issues (Fig. 13). Again, I have highlighted the
smaller ellipses. The mass of overlapping ellipses for the earlier issues in the top-left
quadrant is unsurprising. This mass of points represents early issues, all struck before the
closing date of the earliest hoard. These issues have had time to circulate, and their
distribution across hoards is largely random variation. Once we move beyond that initial
mass, we see RRC 286/1 once again lying in ‘the gap’. Contrast this with RRC 271 where the
centre of the ellipse is in ‘the gap’, but the size of the ellipse clearly shows how unreliable its
position is, largely because there are only 20 coins in the analysis. Once we move beyond
those issues, however, we can start seeing a sequence in the highlighted issues. For example,
RRC 308 appears to be later in the sequence than RRC 316 or 317, but RRC 318 could either
49 Lockyear 2013; Ringrose 2012.
50 The previously published analysis of this data using bootstrapping (Lockyear 2013, pp. 4747–8) only
contained the 38 hoards used in the earlier analyses (Lockyear 2007, pp. 77–9).
51 I have examined the relationship between sample size and the size of the ellipses elsewhere (Lockyear 2013,
pp. 4751–2) and shown that this varies from data set to data set.
be earlier or the same date. Other rare, unreliable issues with large ellipses include RRC 324
with only nine coins in the data set or RRC 269 with only 17 coins.
For our purposes here, we can be confident in the existence of ‘the gap’, and we can start
to see which hoards are likely to be reliably associated with it, such as Santa Elena, La Oliva
or Sierra Morena (hoards SEL, OLI & CEN). We can also pick out some of the issues that
clearly seem to be associated with ‘the gap’ such as RRC 285, 289, 291 and 296. The results
suggest some adjustments are needed to the relative sequence of issues suggested by
Crawford and Mattingly, but further work would be needed to develop a chronological
ordering of all the issues.
Can we divide the issues into groups associated with before, during and after ‘the gap’?
It would be helpful to know which issues are associated with ‘the gap’ and to see how they
are represented in the hoards, much as we divided the hoards into early, middle and later
period hoards as a result of the Cluster Analysis. Doing this in a repeatable, objective manner
without using the rival dating schemes is problematic. One final technique would allow us to
do this at the level of issues: Hierarchical Clustering of Principal Components, or HCPC.52
We have already seen how Cluster Analysis creates groups, and how Correspondence
Analysis identifies the main patterns while discarding the ‘noise’. HCPC combines these two
techniques by performing a Cluster Analysis not on the original data, but on the scores
derived from the CA. It does this because the CA has discarded the ‘noise’. The results can
be directly plotted on a CA map.
Applying the technique to our data suggests three groups in the hoards, and three groups
in the issues. Examining the grouping of the hoards (Fig 14), it is very similar to that we
obtained using Dmax-based clustering earlier. Excluding the two singletons from that
analysis, eight hoards are grouped differently in this analysis.53 Some of those hoards are
quite small, e.g., the Mértola hoard (MER) has only 21 coins in this analysis, or the Manfria
hoard (MNF) discussed above. Remembering that we are dissecting a continuum and not
looking for clearly defined groups, this is a very good result. I shall refer to these hoards as
being from the early, middle and late clusters. It should be remembered that this is not the
same as having an early, middle or late closing date as this is based on the overall pattern in
the hoards, not the specifics of the closing date.
Figure 14: HCPC clusters plotted on the CA map of the hoards, first and second axes of
inertia.
Examining the issues (Fig. 15), we can see that the three groups divide nicely on the map.
RRC 286/1 is placed in the early cluster which, remembering Fig. 11, makes sense. We now
have a way of looking at the issues in three groups which we can interpret as roughly
correlating to early, middle and late issues. As with the hoards I will refer to these as early,
middle and late clusters. The hoards and issues which make-up these clusters is given in
Table 5.
52 Alberti 2013, pp. 40–2, 47–50.
53 Almadenejos (ALM), Córdoba (CO2), Cástulo (CSL), Mértola (MER), Manfria (MNF), Penhagarcía (PNH), Ricina
(RCN), Taranto (TR1).
Figure 15: HCPC clusters plotted on the CA map of the issues, first and second axes of
inertia.
Tables 2a and 2c show the number of coins from each issue in hoards of each cluster, i.e.,
coin types in the early cluster in hoards of the early cluster, and so on. The percentages in
Tables 2b and 2d are calculated from all the coins in hoards of that cluster. From these tables
we can see that early coins in early hoards are evenly spread between Italy and Iberia, middle
coins in middle hoards are, however, dominated by Iberia (25.3% compared to 5.3%)
whereas the latest coins are again associated with Italy. Even in terms of just raw numbers,
Iberia dominates the middle period hoards with 2,778 coins compared to only 761 from
Italy.54
We can check the influence of the hoards which do not fit the map of the first two axes
from CA by excluding all hoards and issues with a low quality. In Tables 2c and 2d I have
excluded any issue or hoard which has a quality of less than 150 on the CA map of the first
two axes. We can see that the pattern is clearer with 23.9% middle cluster coins coming from
middle cluster Iberian hoards but 32.1% of late cluster coins coming from late cluster Italian
hoards.
We have now successfully defined which issues are associated with which part of our
hoard pattern, and these are listed in Table 5.
54 Remember these are well-identified denarii from issue 260/1 onwards.
3 Discussion
To summarise the results so far:
1. The bar charts show that the previously identified pattern in closing dates can be seen
whichever dating scheme is chosen.
2. The cluster analyses show that this pattern persists through the date structure of the
hoards, again no matter which scheme is chosen.
3. The CA and bootstrapped CA shows that the pattern can also be identified at the level
of individual issues. The pattern can be seen to be ‘real’ although the association of
some issues or hoards needs to be treated with care, mainly in the case of small hoards
or small issues.
4. The HCPC allows us to group the issues into those mainly associated with the early
period (RRC 260–284, 286, 289), the middle period (RRC 285, 287, 290–298, 300–
307, 311–314) and late period (RRC 299, 308, 316–330). Only a small number of
issues appear out of sequence compared to RRC, e.g., 299.
There are two levels of interpretation needed to explain the observed pattern. The first level is
to explain the mechanical processes by which the pattern was created. The second level is to
explain why those mechanical processes occurred.
Possibilities for the first level are:
1. Variation in the pattern of coin production.
2. Variation in the pattern of coin supply.
3. Variation in the numbers of hoards buried.
4. Variation in the numbers of hoards recovered in Antiquity.
5. Variation in the numbers of hoards recovered/published in modern times.
These possibilities are not necessarily exclusive. To deal with option 5 first, I cannot see how
selective recovery and reporting could have created the observed pattern. In fact, it is because
of the efforts of Crawford as well as Italian and Spanish scholars that we have enough data to
undertake an analysis like this.55
Option 1 can be examined by looking at the relative size of the issues (Fig. 16).56
Although the pattern does vary from year to year, there are no obvious breaks in production.
All the issues in this period are thought to have been struck in Rome with the one exception
of RRC 282 which was struck at Narbo. I am unaware of anyone who has argued that any of
these issues were struck in Spain.
Figure 16: Size of issues defined by RISC values for the period 118–100 BC (RRC
chronology).
As regards option 2, clearly the pattern of supply is a factor here. For a short period of
time, probably about 10 years, the Roman state appears to have been preferentially sending
55 A comparison could be drawn with hoard pattern between Romania and Bulgaria which is an artefact of
differential reporting (Lockyear 2007, pp. 167–8). Thanks to the efforts of Evgeni Paunov, this situation is
improving.
56 I coined the term ‘relative issue size coefficient’ as an alternative to treating Crawford’s die estimates as
actual counts (Lockyear 1999, p. 226).
new coin to Iberia rather than spending it within the Italian peninsula. Keeping in mind that
hoards cannot be dated to periods where no coins were supplied, some of our ‘early’ Italian
hoards may have been deposited in ‘the gap’ but contain no coins of that period. Similarly,
some of the Iberian hoards which close in ‘the gap’ were probably deposited after it. As
shown in the cluster analysis, the Olival da Soalheira and São João dos Caldeireiros hoards
(OSO & SJ1) have a late closing date, but an overall structure similar to hoards from ‘the gap’,
which supports this idea. It also seems likely that there was an increase in hoarding activity
and non-recovery within Iberia (options 3–4) at this time.
How might we interpret this pattern of supply and hoarding? Crawford57 showed that the
numbers of hoards recovered reflected periods of violence but Kent58 demonstrated that the
location of those hoards did not reflect the location of that violence. Duyrat59 compared the
distribution of hoards in Hellenistic Syria between the 4th and 1st centuries BC with periods
on known warfare, and found that increases in hoarding were associated with some events but
not others. In a recent paper, Guest60 has reviewed the history of linking hoards to military
activity or periods of unrest, and also discussed the problems with using this explanation
universally. For example, the mass of radiate hoards found in Britain in the second half of the
3rd century AD cannot be easily explained in this way.61
Examining the broader pattern of hoarding (Fig. 17), we can see that there are very few
gaps in the sequence in Italy except at those times when very few coins were struck, e.g., the
gap in the 60s BC. The peaks in Italy in the 80s, 70s and 40s are connected to the Social and
Civil Wars and Spartacus’ revolt. They do not just reflect increased hoarding activity,
however, but are also a reflection of the pattern of coinage production.62 The peak of hoards
in the 70s, for example, is partly the result of the small size of coin issues in the 60s BC.
Figure 17: Numbers of hoards with over 20 well identified coins in the CHRR database from
the late Republic classified according to date of latest coin following RRC. (Data as of
July 2015.)
The pattern of hoarding in Iberia, however, is very erratic when compared to Italy. There
is a large peak in the period we are examining, but the numbers of hoards from Iberia is never
greater than Italy at other dates. Most curiously, there are no peaks in hoards in the 130s BC
and the 80s BC when we know there was military activity in Iberia.
Although I agree with Guest that the universal application of the unrest/military model for
the explanation of hoarding activity is flawed, in this case we are not only dealing with an
increase in hoarding, but the deliberate supply of coins to Iberia by the Roman state and not
to Italy. This pattern of supply over a relatively short period of ten years makes it unlikely
that longer term processes such as ritual activity are the cause. This pattern of supply may be
linked to issues of taxation, for example, but the simplest explanation is that the Roman State
was having to pay for something, possibly the military. Conversely, we have no historical
references to military activity in Iberia in this period apart from the incursion of the Cimbri
57 Crawford 1969a.
58 Kent 1988.
59 Duyrat 2011. I would like to thank David Biedermann for bringing my attention to this paper and Frédérique
Duyrat for supplying me with a copy.
60 Guest 2015.
61 Bland 2018.
62 Lockyear 2012, pp. 203–7.
and Teutones into northern Spain in 103 BC from whence they were pushed back into Gaul
by the Celtiberians before their eventual defeat by Marius in 102 and 101 BC. Erdkamp63 has
demonstrated how difficult it is to supply an army on campaign in Iberia. Maintaining a
military presence when not on campaign must have required a steady supply of cash.64 This
could be the explanation for the supply of coinage, but what about the increase in hoards?
Villaronga attributes part of the pattern to the incursion of the Cimbri and Teutones,65 but
many of our hoards are a long way to the south. Remembering the famous incident recorded
by Pepys in his diary, mere rumour could be sufficient to cause hoards to be buried at some
distance from the location of the troubles.66
This purely military explanation for the pattern is at odds, however, with the distribution
of hoards compared to recorded military events elsewhere. We know that Rome was at war
with Jugurtha in North Africa from 112–106 BC, and that successive armies were stationed in
southern Gaul to fight the Cimbri and Teutones including that defeated at the Battle of
Arausio (modern Orange) in 105 BC and the eventual victories at Aquae Sextiae (Aix-en-
Provence) in 102 BC and Vercelli in 101 BC. We have, however, not one hoard of
Republican denarii from North Africa or southern Gaul at this date. Perhaps this is a
reflection of the wider adoption of the denarius in Iberia compared to those regions.
Rather than the pattern representing an increased level of military activity — not
necessarily actual campaigning — at the end of the 2nd century BC in Iberia, perhaps it is a
result of a need for the Roman state to both obtain fresh manpower? The defeat at Arausio is
claimed to have resulted in 80,000 casualties. Or maybe simply to avoid having to open a
third front at this period? Whatever the correct explanation, this could provide a context for
the pattern of coin finds in Italy at this period discussed by Stannard67 which appears to show
strong links between Baetica and Italy at the end of the second century.
In this paper I have conclusively demonstrated that ‘the gap’ exists, and is also reflected
in the pattern of supply of coinage. Although the precise dating of ‘the gap’ remains
uncertain, it is not a product of the dating schemes used. The interpretation of the pattern,
however, remains much more speculative and open to other possible explanations. In
achieving these results, I have also demonstrated the strength and applicability of multivariate
statistical analysis to coin data. Undertaking these analyses is relatively straightforward using
the freely available R statistical system.68 More time-consuming is the collection of the data.
For the Roman Republic, however, this is now freely available via the CHRROnline
database.69
Appendix A: Hoard details
The format of the catalogue is as follows: the name of the hoard (as in CHRR) is followed by
the date it was found, if known, and then the hoard’s identifier is given in SMALL CAPITALS.
The country where the hoard was found is followed by the closing date using the RRC
chronology. After the RRCH number (if listed), the numbers of coins in the hoard is then
63 Erdkamp 2010.
64 Chaves Tristán 2014.
65 Villaronga 1993.
66 Wheatley 1895, vol. 6, p. 379, vol. 7, pp. 143f.
67 Stannard 2005.
68 R Development Core Team 2012.
69 http://numismatics.org/chrr/; Lockyear 2016.
given, followed by the number of well-identified denarii used in the analysis in [ ]. The
source of the data is then given: MHCPR for Michael Crawford’s personal records held in the
British Museum and PUB for a published source. If there are any comments about the find,
these are then given followed by the references. An * indicates the source of the data in the
CHRR database.
La Loba (LOB)
Spain, 119 BC
13 denarii [13]
PUB
Found in the wall of a house during the
excavation of the mine ‘La Loba’.
No associated artefacts.
*Chaves Tristán (1996), pp. 51–54
Patrica 1935 (PTA)
Italy, 119 BC
100 denarii [100]
PUB
Found during agricultural work in a ceramic
vessel and was divided amongst the finders.
Only part of the hoard was recovered.
*Backendorf (1998), pp. 99, 392–3; Petrillo
(1990).
Gerenzago 1909 (GER)
Italy, 118 BC
RRCH 167
60 denarii, 6 victoriati, 2 quinarii, 54 copies
of drachmae of Massalia [49]
MHCPR
Found in a ceramic vessel.
Backendorf (1998), pp. 72, 333; Ricci (1909);
Patroni (1909).
Jesi (JES)
Italy, 118 BC.
67 denarii of an unknown total [67]
PUB
Backendorf (1998), pp. 77–8, 341–2; *Sorda
(1973).
Naples (NAP)
Italy, 118 BC.
8 denarii [8]
MHCPR
Terranova di Sicilia (TDS)
Sicily, 118 BC
RRCH 168
78 denarii [71]
MHCPR
Bevagna 1929 (BEV)
Italy, 117 BC
RRCH 171
784 denarii, 5 sestertii, 104 victoriati [721]
MHCPR
Found in a ceramic vessel during work on a
sports field.
Backendorf (1998), p. 42; Catalli (1991, p.
336).
Lauterach 1880 (LAU)
Austria, 117 BC
RRCH 170
23 denarii, 1 plated hybrid, 3 Celtic silver
coins [22]
MHCPR
With silver and bronze ornaments.
Lockyear (2007), p. 249.
Laives Reif 1985 (LVS)
Italy, 116 BC
22 denarii [22]
PUB
Complete. Found during the stripping of a site
before excavation with a bulldozer.
*Backendorf (1998), pp. 78, 342; Vittorio
(1990).
Maddaloni 1913 (MAD)
Italy, 116 BC
RRCH 172
310 denarii (+25 extraneous examples), 26
victorati, (+1 extraneous misc. bronze) [283]
Found during the construction of a pipeline in
a coarse pottery jar.
MHCPR
Maiuri (1914); Backendorf (1998), pp. 80–1,
346–8.
Naples, National Museum (MNP)
Italy, 116 BC
35 denarii [35]
PUB
Found in museum collection, unsure origin
and unsure it is a hoard but looks like it from
the composition.
*Giove (2002).
Chiclana de Segura 1972 (CHL)
Spain, 115 BC
22 denarii [20]
PUB
Found with 38 pieces of jewellery.
*Chaves Tristán (1996), pp. 58–60.
Carissa 1920 (CRS)
Spain, 115 BC
4 denarii, 73 Iberian denarii [4]
PUB
*Chaves Tristán (1996), pp. 61–5.
Pozoblanco, before 1870 (PZ1)
Spain 115 BC
RRCH 174
84 denarii out of 1000 [79]
MHCPR
Chaves Tristán (1996), pp. 55–7; Villaronga
(1993), no. 83.
Taranto 1897 (TR1)
Italy, 114 BC
RRCH 176
102 denarii [96]
MHCPR
Backendorf (1998), pp. 123, 450; Gabrici
(1898); Mattingly (2004), hoard Ta.
Villanueva de Córdoba 1958 (CO1)
Spain, 113 BC
130 denarii (although 1 is missing) [127]
MHCPR
Found with two bracelets and a bronze situla.
Chaves Tristán (1996), pp. 66–79; Mattingly
(2004), hoard C2; Villaronga (1993), no. 89.
Herdade da Mìlia (HDM)
Portugal, 113 BC
492 denarii (+16 extraneous denarii) [473]
PUB
Found in a ditch in an amphora.
Chaves Tristán 1999, pp. 166–8; *Marques de
Faria 1991–2 pp. 80–3.
Alcalá del Río 1979 (ADR)
Spain, 112 BC
163 denarii [158]
PUB
Blázquez (1987–8), no. 27; *Chaves Tristán
(1996), pp. 79–92; Villaronga (1993), no. 70.
Bugiuleşti 1963 (BUG)
Romania, 112 BC
RRCH 177
2 denarii [2]
PUB
Listed by Chiţescu as isolated finds.
*Chiţescu (1981), pp. 335–6.
La Barroca 1953 (LAB)
Spain, 112 BC
RRCH 178
74 denarii, 2 Iberian denarii, 42 drachmae of
Emporiae [69]
MHCPR
Blázquez (1987–8), No. 25.
Mértola (MER)
Portugal, 112 BC
53 denarii [53]
PUB
Chaves Tristán 1999, pp. 167–8; *Marques de
Faria 1991–2 pp. 84–6.
Segaró 1881 (SEG)
Spain, 112 BC
RRCH 180
47 denarii, 2 Iberian denarii, 963 drachmae of
Emporiae (+3 extraneous bronze coins) [43]
MHCPR
IGCH 2342 lists this hoard as S. Agarro,
Mommsen (1881, p. 809) lists it as Segaró,
Villaronga (1993), No. 47 as Segueró.
El Centenillo 1911 (EL1)
Spain, 110 BC
RRCH 181
72 denarii (+1 extraneous denarius of 318/1b),
2 victoriati [71]
PUB
“The coins and other objects, which came to
light on digging the foundation for a hut, were
scattered in the soil, and if they had originally
been gathered altogether in a receptacle all
traces of it had disappeared… The hoard
consisted of… the armlet and the fragments of
a torc…, of fragments of silver ear-rings and
of other ornaments” (Hill and Sanders 1912, p.
65).
Hill and Sanders (1912), *Chaves Tristán
(1996), pp. 194–7, Marcos Alonso (2009).
Córdoba 1916 [?1915] (CO2)
Spain, 109 BC
RRCH 184
224 denarii, 1 victoriatus, 80 Iberian denarii, 1
drachm of Suguntum [214]
PUB
Found during the digging of a hole for a large
oil-vessel. No surviving container but burial
site is within a Roman cemetery either side of
the Roman road to Cástulo (Hildburgh 1922,
pp. 161–2). “The present find included, in
addition to the bowl, the coins, and the rough
lumps above mentioned, a fragment of a
second vessel and a number of interesting
personal ornaments of several kinds, some
bent out of shape, some with small parts
missing.” (Hildburgh 1922, p. 163).
Chaves Tristán (1996), pp. 93–104; *Crawford
(1969b), pp. 85–93; Hildburgh (1922);
Mattingly (1925), pp. 395–6; Mattingly
(2004), hoard C1.
Baix Llobregat (LLO)
Spain, 109 BC
117 denarii [112]
PUB
Blázquez (1987–8), no. 28; *Villaronga
(1977); Villaronga (1993), no. 49.
Montoro 1936 (MON)
Spain, 109 BC
RRCH 182
20 denarii [15]
MHCPR
Blázquez (1987–8), no. 29; Chaves Tristán
(1996), pp. 378–9; Villaronga (1993), no. 79.
Strongoli 1931 (STR)
Italy, 109 BC
RRCH 183
4 denarii, 3 asses, 1 quadrans [4]
MHCPR
Backendorf (1998), pp. 118–9.
Sarrià 1870 (SAR)
Spain, 108 BC
50 denarii [48]
PUB
Error in publication/CHRR corrected
16/07/2015.
Probably out of 400 denarii. The hoard was
found during works and divided between the
finder’s eight children. This appears to be one
lot of the eight.
Blázquez (1987–8), no. 32; *Villaronga
(1982); Villaronga (1993), no. 50.
Albánchez de Úbeda 1979 (ADU)
Spain, 106 BC
16 denarii [15]
PUB
Blázquez (1987–8), no. 33; Chaves Tristán
(1996), pp. 127–37; *de Paula Perez Sindreu
(1984).
Sierra Morena (El Centenillo) 1920 (CEN)
Spain, 106 BC
RRCH 186
617 denarii [592]
PUB
Blázquez (1987–8), no. 34; *Chaves Tristán
(1996), pp. 120–6;
Mattingly (2004), hoard SM;
Sandars (1921); Villaronga (1993), no. 87.
Pozoblanco 1980s (PZ3)
Spain, 106 BC
74 denarii [74]
PUB
*Chaves Tristán (1999), pp. 144–50.
Castillo de las Guardas 1856 (CLG)
Spain, 105 BC
113 denarii [113]
PUB
Found with other silver objects.
Blázquez (1987–8), no. 35; *Chaves Tristán
(1996), pp. 178–88.
Jaén (JA2)
Spain, 105 BC
128 denarii [128]
PUB
*Chaves Tristán (1999), pp. 151–61.
Torre de Juan Abad 1934 (JUA)
Spain, 105 BC
RRCH 189
476 denarii (+2 extraneous denarii) [476]
PUB
In a silver bowl with silver ornaments.
*Vidal Bardán (1982); Blázquez (1987–8), no.
37; Chaves Tristán (1996), pp. 138–76;
Mattingly (2004), hoard TJ; Villaronga (1993),
no. 88.
Avvocata (AVV)
Italy, 104 BC
RRCH 190
25 denarii [21]
MHCPR
Backendorf (1998), p. 40.
Aznalcóllar 1908–1919? (AZN)
Spain, 104 BC
35 denarii [35]
PUB
Blázquez (1987–8), no. 39; Chaves Tristán
(1996), pp. 189–93; *Crawford (1984),
Villaronga (1993), no. 71.
Cogollos de Guadix 1958 (COG)
Spain, 104 BC
83 denarii [83]
PUB
Blázquez (1987–8), no. 40; Chaves Tristán
(1996), pp. 198–206; *Mendoza Eguaras
(1978); Villaronga (1993), no. 75.
Penhagarcía, about 1920 (PNH)
Portugal, 104 BC
RRCH 191
110 denarii [103]
MHCPR
Out of a larger hoard, with gold and silver
ornaments.
Blázquez (1987–8), no. 38; Villaronga (1993),
no. 82.
Manfria, before 1955 (MNF)
Sicily, 103 BC
RRCH 198
33 denarii [32]
PUB
Crawford (1974, p. 67, note to table XI) states
that this hoard is part of the Terranova de
Sicilia hoard (TDS) but I have been unable to
locate the revised listing.
*Griffo (1958)
La Oliva 1848 (OL2)
Spain, 103 BC
45 denarii [45]
PUB
Chaves Tristán (1996), p. 366 argues that this
hoard is not part of OLI (RRCH 197) as
believed by Crawford. Mattingly (2004)
combines the two hoards following Crawford.
Hoard was originally 600–700 coins.
*Chaves Tristán (1996), pp. 366–8; Mattingly
(2004), hoard Oli.
Puebla de los Infantes, around 1985 (PUE)
Spain, 103 BC
145 denarii, 2 quinarii, 2 victoriati, 1
quadrigatus [131]
PUB
Villaronga (1993, pp. 45–6) argues that the
poor condition of the coins indicates that they
are the unsaleable parts of a number of
different hoards.
Blázquez (1987–8), no. 54; *Chaves Tristán
(1988); Chaves Tristán (1996), pp. 214–29,
Villaronga (1993), no. 84.
Santa Elena 1936 (ELN)
Spain, 102 BC
21 denarii [21]
PUB
Blázquez (1987–8), no. 53; *Chaves Tristán
(1996), pp. 235–7; Mattingly (2004), hoard
SE; Villaronga (1993), no. 86.
San Lorenzo del Vallo 1950 (LOR)
Italy, 102 BC
RRCH 195
310 denarii, 1 copy of a denarius [299]
MHCPR
Found during agricultural work.
Backendorf (1998), p. 112; Generoso (1990);
Mattingly (2004), hoard SL; Procopio (1952).
Largo di Torre Argentina, Rome (LTA)
Italy, 102 BC
16 denarii, 3 quinarii [15]
MHCPR
Found during excavations of the temples.
Pagliuzza 1989–1991 (PAG)
Sicily, 102 BC
390 denarii [390]
PUB
Out of 541 denarii. First lot of 403 coins
found during agricultural works, second lot of
138 found during excavations of the rural site
in the following years. The 390 coins
displayed at the exhibition in 1991 are listed
and include examples of all issues. The
remaining 151 examples were being cleaned
and were to be published by R. Macaluso.
Found with sherds of a ceramic vessel.
Mattingly (2004), hoard Pa; Mantegna
Pancucci et al. (1991).
Rio Tinto, between 1903 and 1910 (RIO)
Spain, 102 BC
RRCH 194
44 denarii [44]
MHCPR
Carson (1952); Chaves Tristán (1996), pp.
230–33; Villaronga (1993), no. 85.
Sierra Morena 1929 (SMR)
Spain, 102 BC
RRCH 196
3 denarii, 9 Iberian denarii [3]
MHCPR
Blázquez (1987–8), no. 43; Chaves Tristán
(1996), pp. 234, Villaronga (1993), no. 65.
Since this study was completed the paper by
Marcos Alonso (2009) has come to my
attention which lists 12 denarii from this hoard
and changes the closing date to 109–108 BC.
Cachapets (CAC)
Spain, 101 BC
266 denarii, 2 victoriati [262]
PUB
*González Prats and Abascal Palazón (1989);
Villaronga (1993), no. 72.
Chão de Lamas (CDL)
Portugal, 101 BC
6 denarii, 1 Iberian denarius [5]
PUB
*Alfaro Asins (1989).
Cástulo 1978 (CSL)
Spain, 101 BC
47 denarii [47]
PUB
Probably complete, no surviving container.
Blázquez (1987–8), no. 46; Chaves Tristán
(1996), pp. 238–43; *Collantes Pérez-Ardá
(1978); Villaronga (1993), no. 74.
Elena (ELE)
Italy, 101 BC
RRCH 199
62 denarii [59]
MHCPR
Out of an unknown larger number. Part of the
hoard was stolen.
Backendorf (1998), p. 65; Pozzi (1960–61), p.
153.
La Oliva 1861 (OLI)
Spain, 101 BC
RRCH 197
1269 denarii, 1 Iberian denarius [1247]
PUB
Chaves Tristán (1996), p. 244 argues that this
hoard is not the same as OL2 contra RRCH
197. Mattingly (2004) follows RRCH and
combines the two hoards.
Blázquez (1987–8), no. 44; *Chaves Tristán
(1996), pp. 244–340; Mattingly (2004), hoard
Oli; Mommsen (1863), pp. 15–64; Villaronga
(1993), no. 80.
Ricina, before 1882 (RCN)
Italy, 101 BC
RRCH 201
299 denarii [271]
PUB
Backendorf (1998, p. 129) lists the hoard
under Villa Potenza. Hoard contained 300
coins.
Backendorf (1998), p. 129, 461–3; Mattingly
(2004), hoard Ri; *Tambroni Armaroli (1882–
4).
Santa Elena 1903 (SEL)
Spain, 101 BC
RRCH 193
568 denarii, 6 Iberian denarii [537]
PUB
Found in two lead purses, with a silver bar.
Blázquez (1987–8), no. 41; Chaves Tristán
(1996), pp. 207–13; *Sandars (1905);
Villaronga (1993), no. 86.
Almadenejos 1976 (ALM)
Spain, 100 BC
100 denarii, 1 victoriatus, 2 Iberian denarii
[95]
PUB
Out of 200 or 300 coins in total. Found with a
torque, a bracelet and a vessel, all in silver.
The latter was destroyed.
*Chaves Tristán (1996), pp. 344–54, 610.
Azuara 1890 (AZR)
Spain, 100 BC
RRCH 204
1 denarius, 409 Iberian denarii [1]
PUB
Blázquez (1987–8), no. 47, *RRCH;
Villaronga (1993), no. 93.
Bologna 1886 (BLG)
Italy, 100 BC
92 denarii [92]
PUB
Out of about 122 coins. Found during building
work.
*Backendorf (1998), 44, 258–9.
Bevagna 1980 (BVG)
Italy, 100 BC
232 denarii, 1 victoriatus, 1 quinarius [227]
PUB
Complete. Found during archaeological
excavation, in a ceramic vessel, which may
have been placed in a wooden box.
*Backendorf (1998), 42–3, 256–8, Catalli
(1991); Mattingly (2004), hoard Be.
Cerignola (CG2)
Italy, 100 BC
97 denarii, 1 victoriatus [96]
MHCPR
Crevillente, before 1949 (CRE)
Spain, 100 BC
RRCH 206
4 denarii, 1 Iberian denarius [4]
MHCPR
Out of 22 coins. Found in a quarry six meters
below the ground surface.
Blázquez (1987–8), no. 49. Mateu y Llopis
(1951), p. 227; Villaronga (1993), no. 57.
Crognaleto 1900 (CRG)
Italy, 100 BC
RRCH 212
149 denarii, 14 quinarii, 4 victoriati [137]
MHCPR
Found by a farmer along the river Vomano.
Backendorf (1998), pp. 64, 315–7; Mattingly
(2004), hoard Cr.
Carovigno 1882 (CVG)
Italy, 100 BC
RRCH 208
478 denarii [459]
PUB
Found during roadworks outside the town in a
ceramic vessel. Out of about 3,000 coins.
*Backendorf (1998), pp. 52, 284–6; Mattingly
(2004), hoard Ca; Nervegna (1882–4).
Gioia dei Marsi 1899 (GDM)
Italy, 100 BC
RRCH 213
222 denarii, 37 quinarii [220]
MHCPR
Found in a bronze vessel. Backendorf (1998,
n. 133), following Fabricotti (1990), believes
that RRCH is in error, and that this hoard dates
much later, to 42 BC, and was found in
Avezzano in 1915. For this article, I have kept
to the listing in Crawford’s personal records
which is based on his inspection of the coins in
the Museo Archeologico di Chieti.
Ambrosoli (1899); Backendorf (1998), pp. 40,
252–3; Fabricotti (1990); Mattingly (2004),
hoard Gi.
Idanha-a-Velha (IAV)
Portugal, 100 BC
1350 denarii (including 4 copies), 12 Iberian
denarii, 5 drachmae of Arse [1340]
PUB
Blázquez (1987–8), no. 52; Mattingly (2004),
hoard Id; *Villaronga (1980); Villaronga
(1993), no. 60.
Imola 1913 (IMO)
Italy, 100 BC
RRCH 210
532 denarii, 12 victoriati [500]
MHCPR
Found during the construction of a
waterworks, in a ceramic vessel. Perhaps sixty
coins went missing.
Backendorf (1998), pp. 74–5, 335–8;
Mattingly (2004), hoard Im; Negrioli (1916).
Olmeneta 1879 (OLM)
Italy, 100 BC
RRCH 203
405 denarii (+ 3 extraneous denarii) [397]
PUB
Found in a ceramic vessel during agricultural
pipe-laying. Backendorf (1998, p. 92)
disagrees with Crawford’s exclusion of 3
denarii and dates this hoard to 85 BC.
Backendorf (1998), pp. 91–2, 372–4; *Fiorelli
(1879); Mattingly (2004), hoard Olm.
Orce 1940 (ORC)
Spain, 100 BC
RRCH 211
73 denarii [72]
MHCPR
Blázquez (1987–8), no. 51; Chaves Tristán
(1996), pp. 355–61, Villaronga (1993), no. 81.
Olival da Soalheira 1896 (OSO)
Portugal, 100 BC
112 denarii [112]
PUB
Found in a grove, the nearest named place is
Monforte da Beira. Found with two gold and
four silver necklaces.
Chaves Tristán (1999), pp. 166–7; *Marques
de Faria (1991–1992), pp. 87–92.
Paterno 1914 (PAT)
Sicily, 100 BC
RRCH 207
150 denarii [149]
May be from a larger hoard.
MHCPR
Salvacañete c.1930 (SAL)
Spain, 100 BC
RRCH 205
12 denarii, 62 Iberian denarii [9]
MHCPR
Found with silver ornaments.
Blázquez (1987–8), no. 48; Villaronga (1993),
no. 63.
São João dos Caldeireiros 1941 (SJ1)
Portugal, 100 BC
658 denarii (+4 extraneous denarii), 1
quinarius [657]
MHCPR
Also known as Mértola. As far as I am aware,
this hoard remains unpublished and thus a
definitive listing is not available.
de Castro Hipólito (1960–1961), p. 89, no.
130; Mattingly (2004), hoard Me.
Appendix B: Technical details of statistical techniques used
The majority of the statistical analyses undertaken in this paper used the R statistical system
which is freely available from the internet.70
Cluster Analysis
The calculation of the Kolmorogov-Smirnov distance was undertaken using a custom script
available from the author. The clustering was undertaken using the agnes command from
the cluster package, version 1.15.271 using the option for average linkage.
Correspondence Analysis
The CA was undertaken using two packages: Canoco, a standalone MS-DOS package72 and
the R package FactoMineR.73 The former was used as the companion program,
Canodraw, creates basic PostScript files which can be manually edited to provide high-
70 R Development Core Team 2012.
71 Maechler et al. 2015.
72 I am grateful to the University of Southampton for allowing me access to their copy of this package.
73 Husson et al. 2015.
quality maps. The latter, via the author’s simSimCA3D, provides the decompositions of
inertia using the format preferred by Greenacre74 which are provided in Tables 3–4.
The final data set (i.e., all hoards over 30 denarii omitting all issues prior to RRC 260 and
RRC 309, as well as Idanha-a-Velha) had a total inertia of 1.119 with the first four axes
having eigenvalues of 0.265, 0.105, 0.074, and 0.059 accounting for 23.7%, 9.4%, 6.5% and
5.3% of the variation in the data set respectively.
The calculation of the scores on the axes by the various CA packages varies slightly and
although the relative positions remain identical the absolute values may vary. Maps 7–10
were created using the results from Canoco. Similarly, the algorithms used may results in an
axis being reversed. This was the case between Canoco and the other packages for the first
axis and so the scores used in the subsequent maps have been reversed to facilitate
comparison.
Bootstrapped CA
Bootstrapped CA is available through Ringrose’s R package cabootcrs.75 Ringrose kindly
sent me his code when the package was under development which I have continued to use as
I am able to manually edit the maps to produce the figures used here. The analysis was run
using 10,000 replicates and Poisson resampling.
HCPC
HCPC is a routine available in the FactoMineR package cited above. The technique first
requires the analysis of the data set using FactoMineR’s CA choosing how many
dimensions should be calculated and thus used in the clustering routine. This was run using
10 dimensions. The clustering is then run on the results of the CA twice, once for the rows
and once for the columns. The dissimilarity measure was Euclidean distance and the
clustering performed using Ward’s method (ward.D). The number of clusters created was
decided by the routine (nb.clust=-1).
The clusters are given in Table 5. The maps were created by editing the earlier maps from
CANODRAW to reflect the cluster memberships.
74 Greenacre 2007.
75 Ringrose 2012.
References
ALBERTI, G. 2013. ‘An R script to facilitate Correspondence Analysis. a guide to the use and
the interpretation of results from an archaeological perspective.’ Archeologia e
Calcolatori 24: 25–53.
ALFARO ASINS, C. 1989. ‘Los denarios del tesoro de Chao de Lamas.’ Gaceta Numismatica
94–5: 131–134.
AMBROSOLI, S. 1899. ‘Gioia dei Marsi.’ Notizie degli Scavi pp. 146–7.
BACKENDORF, D. 1998. Römische Münzschätze des zweiten und ersten Jahrhunderts v. Chr.
vom italienischen Festland. Studien zu Fundmünzen der Antike, Band 13. Gebr. Mann
Verlag, Berlin.
BAXTER, M. J. 1994. Exploratory Multivariate Analysis in Archaeology. Edinburgh
University Press, Edinburgh.
BAXTER, M. J. 2003. Statistics in Archaeology. Arnold, London.
BAXTER, M. J. & H. E. M. COOL 2010. ‘Correspondence Analysis in R for archaeologists: an
educational account.’ Archeologia e Calcolatori 21: 211–228.
BLAND, R. 2018. Coin hoards and Hoarding in Roman Britain AD 43–c.498. British
Numismatic Society Special Publication No. 13. Spink, London.
BLÁZQUEZ, C. 1987–1988. ‘Tesorillos de moneda republicana en la península Ibérica.
Addenda a Roman Republican Coin Hoards.’ Acta Numismática 17–18: 105–142.
CARSON, R. A. G. 1952. ‘A hoard of Roman Republican denarii from Rio Tinto, Spain.’
Numismatic Chronicle, Sixth series 12: 129–30
CATALLI, F. 1991. ‘Un ripostigio monetale da Bevagna.’ In R. Martini & N. Vismara (eds.),
Ermanno A. Arslan Studia Dicata. Parte II. Monetazione romana repubblicana ed
imperiale, Glaux. Collana di Studi e Ricerche di Numismatica 7, pp. 311–58. Ennerre,
Milan.
CHAVES TRISTÁN, F. 1988. ‘Tesoros de monedas inéditas, ocultados en la Bética durante la
República: I. El conjunto de Puebla de los Infantes (Sevilla).’ Anas 1: 133–168.
CHAVES TRISTÁN, F. 1996. Los Tesoros en el Sur de Hispania. Conjuntos de denarios y
objetos de plata durante los siglos II y I a C. Fundación el Monte, Seville.
CHAVES TRISTÁN, F. 1999. ‘De tesoros fragmentados y otros problemas. Nuevos hallazgos de
denarios repubicanos en la Hispania Ulterior.’ In F. Villar & F. Beltrán (eds.), Pueblos,
Lenguas y Escrituras en la Hispania Prerromana, pp. 141–72. Universidad de
Salamanca, Salamanca.
CHAVES TRISTÁN, F. 2014. ‘Monedas: entre la guerra y la paz.’ In F. Cadiou &
M. Navarro Caballero (eds.), La Guerre et ses Traces. Conflits et Sociétés en Hispanie à
l’Époque de la Conquête Romaine (IIIe–Ier s. a.C.), pp. 343–68. Ausonius Éditions,
Bordeaux.
CHIŢESCU, M. 1981. Numismatic Aspects of the Dacian State. British Archaeological
Reports, Oxford. International Series 112.
COLLANTES PÉREZ-ARDÁ, E. 1978. ‘Hallazgo de denarios romanos de Cástulo (Jaén).’
Numisma 28: 19–24.
COOL, H. E. M. & M. J. BAXTER 1995. ‘Finds from the fortress: artefacts, buildings and
correspondence analysis.’ In J. W. Wilcock & K. Lockyear (eds.), Computer
Applications and Quantitative Methods in Archaeology 1993, pp. 177–182. British
Archaeological Reports International Series 598, Oxford.
COOL, H. E. M. & M. J. BAXTER 1999. ‘Peeling the onion: an approach to comparing vessel
glass assemblages.’ Journal of Roman Archaeology 12: 72–100.
COOL, H. E. M. & M. J. BAXTER 2002. ‘Exploring Romano-British finds assemblages.’
Oxford Journal of Archaeology 21(4): 365–80.
CRAWFORD, M. H. 1969a. ‘Coin hoards and the pattern of violence in the late Republic.’
Papers of the British School at Rome 37: 76–81.
CRAWFORD, M. H. 1969b. ‘The financial organization of Republican Spain.’ Numismatic
Chronicle, Seventh series 9: 79–93.
CRAWFORD, M. H. 1969c. Roman Republican Coin Hoards. Royal Numismatic Society,
London. Special Publication No 4.
CRAWFORD, M. H. 1974. Roman Republican Coinage. Cambridge University Press,
Cambridge.
CRAWFORD, M. H. 1984. A catalogue of Roman Republican coins in the collections of the
Royal Scottish Museum, Edinburgh. Royal Scottish Museum, Edinburgh. Art and
Archaeology No. 6.
CRAWFORD, M. H. 1985. Coinage and Money under the Roman Republic. Methuen, London.
DE CASTRO HIPÓLITO, M. 1960–1961. ‘Dos tesouros de moedas romanas em Portugal.’
Conimbriga 2–3: 1–166.
DE PAULA PEREZ SINDREU, F. 1984. ‘Tesorillo de Albánchez (Jaén).’ Gaceta Numismatica
74–5: 107–111.
DEROSE EVANS, J. (ed.) 2013. A companion to the archaeology of the Roman Republic.
Wiley Blackwell, Chichester.
DUYRAT, F. 2011, ‘Guerre et thésaurisation en Syrie hellénistique, IVe-Ier s. av. J.-C.’ In T.
Faucher, M.-C. Marcellesi and O. Picard (eds.), Nomisma: la circulation monétaire dans
le monde grec antique. Actes du colloque internantional, Athènes, 14–17 avril 2010, pp.
417–31. Bulletin de Correspondance Hellénique Supplément 53.
ERDKAMP, P. 2010. ‘Supplying armies in the Iberian Peninsula during the Republic.’ In
C. Carreras & R. Morais (eds.), The Western Roman Atlantic Façade. A study of the
economy and trade in the Mar Exterior from the Republic to the Principate, pp. 135–43.
British Archaeological Reports, Oxford. BAR International Series 2162.
ESTY, W. W. 2003. ‘Statistics in Numismatics.’ In C. Alfaro & A. Burnett (eds.), A Survey of
Numismatic Research 1996–2001, Special Publication No. 14, pp. 921–7. International
Association of Professional Numismatists, Madrid.
FABBRICOTTI, E. 1990. ‘Un tesoretto da Avezzano.’ Bollettino di Numismatica 14–15: 49–
108.
FIORELLI, G. 1879. ‘Olmeneta.’ Notizie degli Scavi pp. 126–9.
GABRICI, E. 1898. ‘Tesoretto di monete repubblicane d’argento.’ Notizie degli Scavi pp. 294–
7.
GENEROSO, C. 1990. ‘Il tesoretto di San Lorenzo del Vallo (Cosenza) 1950.’ Bollettino di
Numismatica 14–15: 109–83.
GIOVE, T. 2002. ‘Napoli, museo archeologico nazionale: tesoretto di denari romano-
repubblicani.’ Annali dell’Instituto Italiano di Numismatica 49: 273–83.
GONZÁLEZ PRATS, A. & J. M. ABASCAL PALAZÓN 1989. El Tesorillo de denarios romanos de
Cachapets (Crevillente – Alincante). Monografías del Museo Arqueológico Municipal
de Crevillente 1. Museo Arqueológico Municipal, Crevillente.
GREENACRE, M. J. 2007. Correspondence Analysis in Practice. Chapman and Hall, Boca
Raton, London and New York, second edition.
GREENACRE, M. J. & J. BLASIUS (eds.) 1994. Correspondence Analysis in the Social
Sciences: Recent Developments and Applications. Academic Press, London and San
Diego.
GRIFFO, P. 1958. ‘Ripostigli della Soprintendenza di Agrigento. II. Denari repubblicani
dall’agro di Gela (Caltanissetta).’ Annali dell’Instituto Italiano di Numismatica 2: 98–
100
GUEST, P. 2015. ‘The burial, loss and recovery of Roman coin hoards in Britain and beyond:
past, present and future.’ In J. Naylor & R. Bland (eds.), Hoarding and the Deposition of
Metalwork from the Bronze Age to the 20th Century: A British Perspective, pp. 101–16.
British Archaeological Reports, Oxford. BAR British Series 615.
HILDBURGH, W. L. 1922. ‘A find of Ibero-Roman silver at Cordova.’ Archaeologia 72: 161–
184.
HILL, G. F. & H. W. SANDARS 1912. ‘Notes on a find of Roman Republican silver coins and
of ornaments from the Centenillo mine, Sierra Morena.’ Numismatic Chronicle, Fourth
series 12: 63–69.
HUSSON, F., J. JOSSE, S. LÊ & J. MAZET 2015. FactoMineR: Multivariate Exploratory Data
Analysis and Data Mining, v. 1.31.3.
KENT, J. P. C. 1988. ‘Interpreting coin finds.’ In P. J. Casey & R. Reece (eds.), Coins and the
Archaeologist, pp. 201–217. Seaby, London, second edition.
KNAPP, R. C. 1987. ‘Spain.’ In M. H. Crawford & A. M. Burnett (eds.), The Coinage of the
Roman World in the Late Republic, pp. 19–37. British Archaeological Reports
International Series No. 326, Oxford.
LOCKYEAR, K. 1995. ‘The supply of Roman Republican denarii to Romania.’ Studii şi
Cercetări de Numismatică 11: 85–102.
LOCKYEAR, K. 1996a. ‘Dmax based cluster analysis and the supply of coinage to Iron Age
Dacia.’ In H. Kamermans & K. Fennema (eds.), Computer Applications and
Quantitative Methods in Archaeology CAA95, pp. 165–178. Institute of Prehistory,
University of Leiden, Leiden. Analecta Praehistorica Leidensia 28.
LOCKYEAR, K. 1996b. Multivariate Money. A statistical analysis of Roman Republican coin
hoards with special reference to material from Romania. Ph.D. thesis, Institute of
Archaeology, University College London.
LOCKYEAR, K. 1999. ‘Hoard structure and coin production in antiquity — an empirical
investigation.’ Numismatic Chronicle 159: 215–243.
LOCKYEAR, K. 2000. ‘Site finds in Roman Britain: a comparison of techniques.’ Oxford
Journal of Archaeology 19(4): 397–423.
LOCKYEAR, K. 2007. Patterns and Process in Late Roman Republican coin hoards 157–2
BC. British Archaeological Reports International Series 1733, Oxford.
LOCKYEAR, K. 2008. ‘Aspects of Roman Republican coins found in late Iron Age Dacia.’ In
V. Spinei & L. Munteanu (eds.), Miscellanea numismatica Antiquitatis. In honorem
septagenarii magistri Virgilii Mihailescu-Bîrliba oblata, Honoraria 4, pp. 147–176.
Editura Academiei Române, Bucureşti.
LOCKYEAR, K. 2012. ‘Dating coins, dating with coins.’ Oxford Journal of Archaeology 31(2):
191–211.
LOCKYEAR, K. 2013. ‘Applying bootstrapped correspondence analysis to archaeological
data.’ Journal of Archaeological Science 40(12): 4744–53.
LOCKYEAR, K. 2016. ‘The Coin Hoards of the Roman Republic Database: The History, the
Data, and the Potential.’ American Journal of Numismatics 18: 157–82.
LOCKYEAR, K. forthcoming. ‘Simplifying complexity’. Paper to be published in Mairat, J.
(ed.) Coin Hoarding in the Roman Empire, Oxford University Press, Oxford. Draft
available at https://www.academia.edu/33343982/Simplifying_Complexity
MAECHLER, M., P. ROUSSEEUW, A. STRUYF, M. HUBERT & K. HORNIK 2015. cluster: Cluster
Analysis Basics and Extensions. R package version 2.0.2.
MAIURI, A. 1914. ‘Maddaloni — Ripostiglio di monete repubblicane scoperte nell’area
dell’antica Calatia.’ Notizie degli Scavi, Fifth series 11: 172–178.
MANTEGNA PANCUCCI, E., D. PANCUCCI & S. VASSALLO 1991. ‘Il ripostiglio monetale e
l’insediamento rurale in localitá Pagliuzza.’ In A. Marotta, C. Greco, F. Spatafora &
S. Vassallo (eds.), Di Terra in Terra. Nuove scoperte archeologiche nella provincia di
Palermo, pp. 141–56. Museo Archeologico Regionale di Palermo, Palermo.
MARCOS ALONSO, C. 2009. ‘El Centenillo (Baños de la Encina, Jaén): Tesoros de 1911
(junio) y 1929.’ In A. Arévalo González (ed.), Actas XIII Congresso Nacional de
Numismática “Moneda y Arqueologia”. Cádiz, 22–24 octubre de 2007, pp. 299–324.
Cádiz and Madrid.
MARQUES DE FARIA, A. 1991–1992. ‘Très tesouros monetários romanos da Época
republicana.’ Nummus 14–15: 79–94.
MATEU Y LLOPIS, F. 1951. ‘Hallazgos monetarios (vi).’ Ampurias 13: 203–55.
MATTINGLY, H. 1925. ‘Some Roman hoards.’ Numismatic Chronicle, Fifth series 5: 395–
401.
MATTINGLY, H. B. 2004. From Coins to History: Selected Numismatic Studies. University of
Michigan Press, Ann Arbor.
MENDOZA EGUARAS, A. 1978. ‘Tesorillo de denarios republicanos de Cogollos de Guadix
(Granada).’ Numisma 28: 25–37.
MOMMSEN, T. 1863. ‘Sopra alcuni ripostigli di denari romani scoperti nella Spagna.’ Annali
dell’Instituto di Corrispondenza Archeologica 35: 5–80. Cazlona pp. 11–15, Oliva 15–
64, Carrara 64–69, Liria 69–75, Cordoba 75–80.
MOMMSEN, T. 1881. ‘Über die antike numismatik Hispaniens.’ Monatsbericht der Königlich
Preussischen Akademie der Wissenschaften zu Berlin. pp. 806–32.
NEGRIOLI, A. 1916. ‘Imola — Tesoretto di monete repubblicane, d’argento scoperto davanti
al palazzo Vescovile.’ Notizie degli Scavi, Fifth series 13: 159–163.
NERVEGNA, G. 1882–1884. ‘Ripostiglio di Consolari in Carovigno.’ Bullettino di
Numismatica e Sfragistica per la Storia d’Italia 1: 281–6.
PATRONI, G. 1909. ‘Gerenzago — Oggetti preistorici e tesoretto di monete d’argento galliche
e romane, trovati presso il castello.’ Notizie degli Scavi, Fifth series 6: 266.
PETRILLO, P. S. 1990. ‘Il tesoretto de patrica.’ Bollettino di Numismatica 14–15: 23–47.
PITTS, M. & D. PERRING 2006. ‘The making of Britain’s first urban landscapes: the case of
late Iron Age and Roman Essex.’ Britannia 37: 189–212.
POZZI, E. 1960–1961. ‘Ripostigli repubblicani romani nel Museo Archeologico Nazionale di
Napoli.’ Annali dell’Instituto Italiano di Numismatica 7–8: 153–245.
PROCOPIO, G. 1952. ‘S. Lorenzo del Vallo (Cosenza) — Ripostiglio di denari repubblicani
romani.’ Notizie degli Scavi, Eighth series 6: 177–184.
R DEVELOPMENT CORE TEAM 2012. R: A language and environment for statistical
computing. R Foundation for Statistical Computing, Vienna.
RICCI, S. 1909. ‘Gerenzago — Ripostiglio di monete galliche.’ Notizie degli Scavi, Fifth
series 6: 299–301.
RINGROSE, T. J. 2012. ‘Bootstrap confidence regions for correspondence analysis.’ Journal of
Statistical Computation and Simulation 82(10): 1397–1413.
RIPOLLÉS, P. P. 2005. ‘Coinage and identity in the Roman Provinces: Spain.’ In C. Howgego,
V. Huechert & A. Burnett (eds.), Coinage and Identity in the Roman Provinces, pp. 79–
93. Oxford University Press, Oxford.
RIPOLLÉS, P. P. 2012. ‘The ancient coinages of the Iberian peninsula.’ In W. E. Metcalf (ed.),
The Oxford Handbook of Greek and Roman Coinage, pp. 356–74. Oxford University
Press, Oxford.
SANDARS, H. 1905. ‘Notes sur un dépôt de monnaies romaines découvert en Espagne
(province de Jaen) en 1903.’ Revue Numismatique, Fourth series 9: 396–405.
SANDARS, H. 1921. ‘Notes on a hoard of Roman denarii found in the Sierra Morena in the
south of Spain.’ Numismatic Chronicle, Fifth series 1: 179–186.
SHENNAN, S. J. 1997. Quantifying Archaeology. Edinburgh University Press, Edinburgh,
second edition.
SORDA, S. 1973. ‘Tesoretto di Jesi.’ Rivista Italiana di Numismatica 75: 111–119.
STANNARD, C. 2005. ‘Numismatic evidence for relations between Spain and central Italy at
the turn of the second and first centuries BC.’ Schweizerische Numismatische
Rundschau 84: 47–80.
TAMBRONI ARMAROLI, E. 1882–1884. ‘Ripostiglio di monete familiari scoperto fra le rovine
dell’antica Ricina nella provincia di Macerata.’ Bullettino di Numismatica e Sfragistica
1: 241–259.
THOMPSON, M., O. MARKHØLM & C. M. KRAAY 1973. An Inventory of Greek Coin Hoards.
American Numismatic Society, New York.
VIDAL BARDÁN, J. M. 1982. ‘Tesorillo de denarios romano-republicanos de Torre de Juan
Abed (Ciudad Real) en el Museo Arqueológico Nacional.’ Acta Numismàtica 12: 79–95.
VILLARONGA, L. 1977. ‘Tesorillo de denarios romanos del Baix llobregat (Barcelona).’ In
XIV Congreso Nacional de Arqueología, pp. 871–6.
VILLARONGA, L. 1980. ‘Tresor de Idanha-a-Velha (Castelo Branco, Portugal) de denaris
romans, ibèrics i drachmes d’Arse.’ Numisma 165–167: 103–117.
VILLARONGA, L. 1982. ‘Nuevo testimonio del paso de los Cimbrios. El tesorillo de Sarrià
(Barcelona).’ Gaceta Numismatica 64: 24–30.
VILLARONGA, L. 1993. Tresors monetaris de la Peninsula Ibèrica anteriors a August:
Repertori i Analisi. Asociación Numismática Española, Societat Catalana d’Estudis
Numismátics, Barcelona.
VITTORIO, A. 1990. ‘Il tesoretto de Laives Reif — via Lichtenstein.’ Bollettino di
Numismatica 14–15: 7–21.
VOLK, T. R. 1994. ‘The hoard as a diagnostic tool. The different patterns to Roman silver
currency in Hispania and Italy at the end of the 2nd century BC.’ In M. Campo (ed.),
Tresors del món antic, pp. 41–62. Museu Nacional d’Art de Catalunya, Barcelona.
WALTON, P. J. 2012. Rethinking Roman Britain: Coinage and Archaeology. Moneta,
Wetteren.
WHEATLEY, H. B. (ed.) 1895. The Diary of Samuel Pepys. George Bell & Sons. Seven
volumes.